Research map / Detection and segmentation / Open-set and few-shot detection
Deep Anomaly Detection Using Geometric Transformations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train a classifier to distinguish geometric transformations and score normality from transformed-image softmax statistics | Image anomaly detection · CIFAR-10 one-vs-all average | AUROC: 64.8 -> 86.0 | Helped |
| Train a classifier to distinguish geometric transformations and score normality from transformed-image softmax statistics | Image anomaly detection · CIFAR-100 one-vs-all average | AUROC: 63.1 -> 78.7 | Helped |
| Train a classifier to distinguish geometric transformations and score normality from transformed-image softmax statistics | Image anomaly detection · Fashion-MNIST one-vs-all average | AUROC: 92.8 -> 93.5 | Helped |
| Train a classifier to distinguish geometric transformations and score normality from transformed-image softmax statistics | Image anomaly detection · CatsVsDogs one-vs-all average | AUROC: 51.6 -> 88.8 | Helped |
| Use a two-headed classifier with the transformation head for OOD scoring | CIFAR-10 OOD detection · Resized Tiny-ImageNet OOD | AUROC: 92.1 -> 95.7 | Helped |
| Use a two-headed classifier with the transformation head for OOD scoring | CIFAR-10 OOD detection · Resized Tiny-ImageNet OOD | AUPR-In: 89.0 -> 96.1 | Helped |
| Use a two-headed classifier with the transformation head for OOD scoring | CIFAR-10 OOD detection · Resized Tiny-ImageNet OOD | AUPR-Out: 93.6 -> 95.4 | Helped |
| Use identity and horizontal flip when the normal digit is invariant to flipping (ablation) | Transformation-based anomaly scoring · MNIST digit 8 normal, digit 3 anomaly | AUROC: 0.646 | Hurt |
| Use identity and horizontal flip when the normal digit is distinguishable under flipping (ablation) | Transformation-based anomaly scoring · MNIST digit 3 normal, digit 8 anomaly | AUROC: 0.957 | Helped |
| Use translation by 7 pixels alongside identity (ablation) | Transformation-based anomaly scoring · MNIST digit 8 normal, digit 3 anomaly | AUROC: 0.919 | Helped |
| Use non-geometric transformations such as blur, sharpening, or gamma correction instead of geometric transformations (ablation) | Transformation-based anomaly scoring · Image anomaly detection | AUROC | Hurt |
| Use the simplified average transformed-class softmax score instead of the Dirichlet score (ablation) | Transformation-based anomaly scoring · Image anomaly detection | AUROC | Hurt |
Closest papers by meaning
- Neural Transformation Learning for Deep Anomaly Detection Beyond Images (ICML 2021 · similarity 0.88)
- Understanding Anomaly Detection with Deep Invertible Networks through Hierarchies of Distributions and Features (NeurIPS 2020 · similarity 0.87)
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.85)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.84)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.84)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.